The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like find, cmp, tar, and test to perform integrity checks on the files being packaged to ensure no broken links or empty files are delivered.
[COMMAND_EXECUTION]: It invokes environment-appropriate commands such as open (macOS), xdg-open (Linux), and explorer.exe (Windows) to reveal the final folder to the user after delivery verification.
[DATA_EXPOSURE]: The skill accesses project files and common user directories (Downloads, Desktop) to aggregate task outcomes into a single reviewable location.
[DATA_EXPOSURE]: It contains explicit defensive instructions in SKILL.md to avoid including secrets, keys, or credentials in the generated package, acknowledging the risk of desktop content being shared.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted conversation data. It mitigates risk by instructing the agent to 'distill' findings into new documents rather than performing raw transcripts or data dumps.
Ingestion points: Conversation history and files specified in SKILL.md (e.g., repo, Downloads).
Capability inventory: Shell execution and file system writes described in SKILL.md for package creation and verification.
Sanitization: Content transformation (distillation) is required by the instructions to ensure the output is a standalone document.
Boundary markers: While explicit delimiters are not mentioned, the distillation logic acts as a significant transformation barrier.

desktop-package